Immediate action needed against request abuse in NIM

NIM performance is starting to degrade, and it’s not hard to see why. A small group of users are hammering the API with constant, high-frequency requests like there’s no shared infrastructure behind it.

Endless loops, poorly configured agents, people trying to brute-force throughput by spamming requests or juggling multiple accounts — that’s not “usage”, that’s just abuse. And on top of that, the same crowd keeps asking for higher RPM limits as if the issue wasn’t obvious.

This isn’t normal usage, and it’s starting to impact everyone else who’s actually using the API properly.

Lowering limits for everyone isn’t the fix, it just hurts normal users. The real issue is how some people are using the API: nonstop requests, high RPS, excessive parallel calls, zero backoff, perfectly timed loops.

Instead of global limits, just slow that down — add small delays, lower their priority, cap concurrent requests. Normal users won’t notice, but spammers will.

Because let’s be honest, if you’re firing requests every second with no pause, you’re the problem, and asking for higher RPM on top of that just makes it worse.